Lukas Haas

SimpleQA Verified, an improved factuality benchmark for LLMs, reveals that Gemini 2.5 Pro achieves state-of-the-art F1-score of 55.6, outperforming GPT-5 on measuring parametric knowledge and hallucination mitigation.

“Co-author of the SimpleQA Verified benchmark paper”

“Co-author of Google Blog post introducing Deep Research Max leveraging Gemini 3.1 Pro for autonomous research.”

Google releases Gemini Deep Research agent via Interactions API with state-of-the-art performance (46.4% on Humanity's Last Exam), along with open-sourced DeepSearchQA benchmark for evaluating multi-step web research tasks.

“Author announcing the release of Gemini Deep Research agent via the Interactions API.”